Introduction of contextual information in a multisensor EKF for autonomous land vehicle positioning

The aim of this article is to define a multisensor extended Kalman filter taking context, modelled by fuzzy subsets, into consideration. A nonlinear state model, based on the vehicle dynamics, and three measurement models (GPS, gyroscopes and accelerometers) are developed. In order to take context into consideration, fuzzy validity bounds of each sensor are defined using x² statistics of the normalized innovation given by each sensor. The fuzzification of the validity domain of each sensor is then introduced in the EKF. Simulations are then realized to evaluate the theoretical results. The proposed method is compared to the case when no contextual information is used and when validity bounds are modelled by classical logic. Results show that the fuzzification leads to an improvement of the vehicle positioning.